summaryrefslogtreecommitdiffstats
path: root/Misc
diff options
context:
space:
mode:
authorTim Peters <tim.peters@gmail.com>2004-08-30 02:44:38 (GMT)
committerTim Peters <tim.peters@gmail.com>2004-08-30 02:44:38 (GMT)
commit47e52ee0c5c8868db903d476b49c3368fce4d79a (patch)
treefed354561308995cb0f8090e201d5138e17ddce8 /Misc
parent48bd7f3a71c05dafec820dbbf76e611add7a9849 (diff)
downloadcpython-47e52ee0c5c8868db903d476b49c3368fce4d79a.zip
cpython-47e52ee0c5c8868db903d476b49c3368fce4d79a.tar.gz
cpython-47e52ee0c5c8868db903d476b49c3368fce4d79a.tar.bz2
SF patch 936813: fast modular exponentiation
This checkin is adapted from part 2 (of 3) of Trevor Perrin's patch set. BACKWARD INCOMPATIBILITY: SHIFT must now be divisible by 5. AFAIK, nobody will care. long_pow() could be complicated to worm around that, if necessary. long_pow(): - BUGFIX: This leaked the base and power when the power was negative (and so the computation delegated to float pow). - Instead of doing right-to-left exponentiation, do left-to-right. This is more efficient for small bases, which is the common case. - In addition, if the exponent is large (more than FIVEARY_CUTOFF digits), precompute [a**i % c for i in range(32)], and go left to right 5 bits at a time. l_divmod(): - The signature changed so that callers who don't want the quotient, or don't want the remainder, can pass NULL in the slot they don't want. This saves them from having to declare a vrbl for unwanted stuff, and remembering to decref it. long_mod(), long_div(), long_classic_div(): - Adjust to new l_divmod() signature, and simplified as a result.
Diffstat (limited to 'Misc')
-rw-r--r--Misc/NEWS14
1 files changed, 13 insertions, 1 deletions
diff --git a/Misc/NEWS b/Misc/NEWS
index 431b343..660c49f 100644
--- a/Misc/NEWS
+++ b/Misc/NEWS
@@ -20,7 +20,11 @@ Core and builtins
to compute 17**1000000 dropped from about 14 seconds to 9 on my box due
to this much. The cutoff for Karatsuba multiplication was raised,
since gradeschool multiplication got quicker, and the cutoff was
- aggressively small regardless.
+ aggressively small regardless. The exponentiation algorithm was switched
+ from right-to-left to left-to-right, which is more efficient for small
+ bases. In addition, if the exponent is large, the algorithm now does
+ 5 bits (instead of 1 bit) at a time. That cut the time to compute
+ 17**1000000 on my box in half again, down to about 4.5 seconds.
- OverflowWarning is no longer generated. PEP 237 scheduled this to
occur in Python 2.3, but since OverflowWarning was disabled by default,
@@ -156,6 +160,14 @@ Tools/Demos
Build
-----
+- Backward incompatibility: longintrepr.h now triggers a compile-time
+ error if SHIFT (the number of bits in a Python long "digit") isn't
+ divisible by 5. This new requirement allows simple code for the new
+ 5-bits-at-a-time long_pow() implementation. If necessary, the
+ restriction could be removed (by complicating long_pow(), or by
+ falling back to the 1-bit-at-a-time algorithm), but there are no
+ plans to do so.
+
- bug #991962: When building with --disable-toolbox-glue on Darwin no
attempt to build Mac-specific modules occurs.